Why the rise of open source AI isn’t hurting Anthropic … yet
TL;DR
Decagon CEO Jesse Zhang argues that mature enterprise use cases are moving from expensive frontier models to smaller open source models, while new use cases still start on top-tier models. Vercel data supports that split: DeepSeek now handles just over a third of tokens on its gateway, while Z.ai’s GLM-5.2 ranks fourth. By spend, Anthropic still accounts for more than half.
Nauti's Take
Anthropic is not losing yet because customers pay for risk reduction at the start of a workflow, not just for tokens. In that phase, the best answer matters more than the cheapest throughput.
Open source then eats the routine work, and that is dangerous enough: if companies move experiments into cheap production faster, the premium share per use case shrinks. Anthropic has to keep owning new problem categories, or premium pricing turns into an expensive first touch.
Briefingshow
The AI economy is splitting by phase: frontier models win uncertain discovery, while cheaper open models take over stable production. For teams, the smart stack becomes dynamic. Use Claude, Opus, or similar models to prove quality and risk first, then cut costs with DeepSeek, GLM, or Nemotron once the workflow is predictable.
Sources